The obvious challenges involved in learning the sounds of a second language (L2) have inspired a long tradition of research and theorizing into how differences in L1 and L2 sound inventories and the corresponding crosslinguistic acoustic relationships shape L2 phonetic and phonological learning. Nevertheless, it is hard to account for the available data without acknowledging that L2 learners seem to learn what L2 categories are present more quickly and easily than one might think, and recent (as well as not so recent) data also indicate strong sensitivity to subtle acoustic detail across the lifespan. This raises interesting questions about the nature of phonetic/phonological categories, which in a L2 have been characterized as imprecise or lacking robustness (e.g., Bordag et al., 2022), as well as about how perception does, or does not, shape L2 category learning. We begin with the insight that learners can leverage variable goodness-of-fit between L2 and L1 categories to gain a foothold on difficult L2 contrasts, implying substantial sensitivity to acoustic detail (e.g., Darcy, et al., 2013). We studied this in late Spanish-English bilinguals with high-intermediate to advanced English proficiency (n = 86, plus 55 L1 English listeners). We compared 5 English vowels, /ɛ/, /æ/, /oʊ/, /ɔ/, /ʌ/, which instantiate some difficult L2 English contrasts. These vowels’ multi-way assimilation to Spanish /e/, /a/, /o/ should provide useful information for learning the underlying categories. We constructed a 72-item continuum encompassing bet-bat-boat-bought-but by varying vowels' F1, F2 (6 steps each), and durations (two steps). Participants rated how well each item matched each target word using a Variable Analog Scale (Apfelbaum et al., 2022). Unlike forced choice, this paradigm is sensitive to both phonetic category structure and acoustic detail. The data were analyzed in two phases, applying various clustering techniques to first identify subgroups of participants with similar results profiles, and then to explore the vowel category knowledge of those subgroups. These methods include archetype analysis and fuzzy C-means clustering; since these are not widely used in language science, one goal of the talk will be to introduce these methods and explore their usefulness. Four participant profiles were identified, and while L1 English and L1 Spanish participants were not evenly distributed among these subgroups, there was a large degree of overlap. Moreover, all subgroups exhibited evidence of 5 underlying vowel categories. While L2 listeners’ categories tended to be more diffuse, gradient response patterns in both groups indicated robust sensitivity to acoustic detail. Thus, apparent difficulty in perceiving L2 contrasts arises from mapping acoustic information to abstract phonetic categories, rather than from reduced sensitivity to acoustic detail or from imprecise category representations.